Why Meta bought Manus — and what it signals for your enterprise AI agent strategy

Facebook and Instagram parent company Meta’s agreement to acquire Manus for more than $2 billion — announced last night by both companies and reported in The Wall Street Journal — marks one of the clearest signals yet that large tech platforms are no longer just competing on model quality, but on who controls the execution layer of AI-powered work.

Manus, a Singapore-based startup founded by Chinese entrepreneurs that debuted earlier this year, has built a general-purpose AI agent designed to autonomously carry out multi-step tasks such as research, analysis, coding, planning, and content generation.

The company will continue operating from Singapore and selling its subscription product while its team and technology are integrated into Meta’s broader AI organization.

Manus co-founder and CEO Xiao Hong, who goes by “Red,” is expected to report to Meta COO Javier Olivan.

The deal arrives as Meta accelerates its AI investments to compete with Google, Microsoft, and OpenAI — and as the industry’s focus shifts from conversational demos to systems that can reliably produce artifacts, complete workflows, and operate with minimal supervision.

Manus as an execution layer, not a chat interface

Manus has consistently positioned itself less as an assistant and more as an execution engine. Rather than answering isolated prompts, its agent is designed to plan tasks, invoke tools, iterate on intermediate outputs, and deliver finished work.

It gained 2 million users on its waitlist alone after unveiling itself in spring 2025. At that time, Manus outperformed OpenAI's Deep Research agent (powered then by the o3 model) and other state-of-the-art systems on the GAIA benchmark, which evaluates how well AI agents complete real-world, multi-step tasks, by more than 10% in some cases.

And in the acquisition announcement last night, Manus said its system has processed more than 147 trillion tokens and created over 80 million virtual computers, metrics that suggest sustained, production-level usage rather than limited experimentation.

Meta, meanwhile, said Manus can independently execute complex tasks such as market research, coding, and data analysis, and confirmed it will continue operating and selling the Manus service while integrating it into Meta AI and other products.

For enterprises, this distinction matters. Many early “agent” systems fail not because the underlying models can’t reason, but because execution breaks down: tools fail silently, intermediate steps drift, or long-running tasks can’t be resumed or audited. Manus’s core value proposition is that it manages those failure modes.

What Manus users were actually doing with the agent

Evidence of that execution-first positioning shows up clearly in Manus’s own community. In the official Manus Discord server, a “Use Case Channel” post shared by a community member named Yesly on March 6, 2025 catalogued real examples of how users were already deploying the agent.

Those use cases went far beyond casual prompting. They included:

• Generating long-form research reports, such as a detailed analysis of climate change impacts on Earth and human society over the next century

• Producing data-driven visual artifacts, including an NBA scoring efficiency four-quadrant chart based on player statistics

• Conducting product and market research, such as comparing every MacBook model across Apple’s history

• Planning and synthesizing complex, multi-country travel itineraries, complete with budget estimates, accommodations, and a generated travel handbook

• Tackling technical and academic tasks, including summarizing high-temperature superconductivity research, proposing PhD research directions, and outlining simulation-based approaches to room-temperature superconductors

• Drafting structured proposals, such as designs for a solar-powered, self-sufficient home with defined geographic coordinates and engineering constraints

Each example was shared as a replayable Manus session, reinforcing that the system wasn’t just generating text, but orchestrating multi-step work to produce finished outputs.

This pattern matters because it shows Manus operating in the messy middle ground where enterprise AI often stalls: tasks that are too complex for a single prompt, but too open-ended for rigid automation.

Manus's recent updates

The pace at which Manus shipped updates was also impressive, which likely added to its momentum with users and as a ripe acquisition target for Meta.

In October, the company released Manus 1.5, an update aimed squarely at where early agent systems tended to break down: long, brittle tasks that lost context or stalled halfway through.

Manus re-architected its core agent engine and saw immediate gains. The company said average task completion times dropped from roughly 15 minutes earlier in the year to under four minutes, nearly a fourfold speedup.

The system dynamically allocated more reasoning time and compute to harder problems instead of treating every task the same. Manus also expanded the agent’s context windows, enabling it to track longer conversations and more intricate workflows without dropping key details. Together, those changes reduced outright task failures and improved output quality for research-heavy, analytical, and multi-step jobs that previously required frequent human intervention.

In December, Manus built on that foundation with version 1.6, extending those execution gains into more autonomous, creative, and platform-spanning work.

The release introduced a higher-performance agent tuned to complete more tasks successfully in a single pass, along with new support for mobile application development, not just web-based projects. Users could describe a mobile app and have the agent handle the end-to-end build process, expanding Manus’s reach beyond the browser. At the same time, the agent carried creative objectives across an entire production arc — from research and ideation to drafting, visual creation, revision, and final delivery — within one continuous session.

That included generating and editing images through a visual interface, assembling presentations and reports, and building full-stack web applications the agent could launch, test, and fix on its own.

Taken together, the updates reinforced Manus’s positioning not as a prompt-driven assistant, but as an execution system designed to stay with a job, adapt when things broke, and reliably deliver finished work across analytical, creative, web, and mobile workflows.

Application-layer traction over proprietary models

Notably, Manus does not train its own frontier model. Reporting on the deal says it relies on third-party AI models from providers including Anthropic and Alibaba, focusing its differentiation on orchestration, reliability, and execution.

That hasn’t prevented commercial traction. Yuchen Jin, co-founder and chief technology officer (CTO) of AI cloud GPU-as-a-service provider Hyperbolic Labs, highlighted this dynamic in a public post discussing the acquisition. Jin noted that Manus by its own admission reached roughly $100 million in annual recurring revenue just eight months after launch, despite having no proprietary large language model (LLM) of its own, relying on the aforementioned providers.

“People keep assuming a small update from OpenAI or Google will wipe out a lot of AI startups,” Jin wrote. “But in reality, the AI application layer should be where most of the opportunity is.”

A similar interpretation came from Dev Shah, lead developer relations at Resemble AI, who argued that Meta didn’t acquire a model company so much as an “environment company” and that “intelligence cannot exist in isolation."

His point? Agentic capability emerges from how models are coupled with tools, memory, and execution environments — a new concept he described as “Situated Agency.”

From that perspective, Manus’s achievement was not training a proprietary foundation model, but engineering an execution layer that allows models like Claude to browse the web, write and run code, manipulate files, and complete multi-step workflows autonomously.

Shah suggested this may align more closely with Meta’s long-term strategy: rather than winning the race for state-of-the-art models, Meta could focus on owning the agentic infrastructure — the orchestration, context engineering, and interfaces — and swap in whichever model performs best over time. If that thesis holds, the Manus acquisition signals a shift toward treating foundation models as interchangeable inputs, while the execution environment becomes the primary source of durable value.

These perspectives help explain Meta’s move. Rather than buying another model team, it is acquiring a system that has already proven it can package existing models into a product users will pay for — and keep using.

What this means for your enterprise AI strategy

For enterprise technical decision-makers, the Manus acquisition is less a vendor endorsement and more a strategic signal.

First, it reinforces that orchestration layers — systems that manage planning, tools, retries, memory, and monitoring — are becoming as important as the models themselves. Enterprises building internal AI capabilities may want to invest more heavily in agent infrastructure that sits above models and can survive rapid shifts in the underlying model ecosystem.

In that sense, building an internal agent layer is not speculative or redundant. It is exactly the class of software that large platforms now view as strategically valuable — whether as acquisition targets or as internal accelerators.

A video recorded ahead of this announcement by VentureBeat founder and CEO Matt Marshall and Red Dragon co-founder Witteveen delves deeper into this subject. Watch it free below or on YouTube.

Second, the deal does not automatically mean enterprises should rush to standardize on Manus itself. Meta’s history with enterprise products gives reason for caution. Tools like Workplace by Facebook gained early adoption but ultimately failed to become deeply embedded enterprise platforms, in part due to shifting internal priorities and inconsistent long-term investment.

That history suggests a measured approach. Enterprises evaluating Manus today may want to treat it as a pilot or adjunct tool, not a foundational dependency, until Meta’s integration strategy becomes clearer.

Key questions include whether Manus remains product-led rather than ad- or data-driven, how governance and compliance evolve under Meta, and whether the roadmap continues to prioritize execution reliability over surface-level integration.

Finally, the acquisition sharpens a broader choice facing enterprises: whether to wait for vendors to define the agent layer, or to build and control it themselves. Manus’s trajectory suggests that the real leverage in AI increasingly lives not in who owns the smartest model, but in who owns the systems that turn reasoning into completed work.

In that light, Meta’s acquisition is less about Manus alone — and more about where the next durable layer of the AI stack is taking shape.

Why this deal matters beyond Meta

From the perspective of some of us here at VentureBeat, the Manus acquisition is best read as confirmation of where value is consolidating in the AI stack (and Meta’s enterprise AI agent ambitions, though the latter is far less assured.)

The defining signal is not that Manus built novel models, but that it demonstrated how quickly well-designed agents can be turned into revenue-generating products by focusing on execution, speed, and concrete outcomes.

That shift — from debating what frontier models can do to measuring what agents actually deliver — increasingly frames how AI progress is evaluated.

The deal also sharpens an important distinction for enterprise readers: this is not primarily about adopting a Meta-backed product, but about recognizing that agent orchestration has become strategically material. Manus succeeded by targeting tractable, real-world tasks and shipping agents that worked end to end, even if those use cases skewed more consumer-oriented.

The broader implication is that enterprises can apply the same approach in their own domains, building agent systems where they already possess data, expertise, and operational leverage.

At the same time, we're cautious about reading this as a direct enterprise buying signal. Meta’s history suggests that long-term enterprise trust is difficult to earn without sustained focus and specialized go-to-market muscle. Where the acquisition may make more immediate sense is on the consumer and small-business side of Meta’s own ecosystem, particularly within products already designed to manage commerce, content, and customer interaction at scale.

Manus’s agentic capabilities map cleanly onto surfaces like Meta Business Suite, where small businesses already juggle content calendars, inboxes, ads, analytics, and monetization tools across Facebook and Instagram. An execution-oriented agent could plausibly automate or coordinate many of those tasks end to end, from drafting and scheduling posts to responding to messages, optimizing ads, or assembling performance reports.

Manus's "Design View" feature, which launched publicly just a week prior to the Meta acquisition announcement and allows users to generate new imagery with editable discrete components using natural language, would seem to be tailor-made for a social network ad creation experience:

Beyond creators and small businesses, those agents could plausibly extend to everyday users navigating Instagram or Facebook for shopping, discovery, or personal expression. An execution-oriented agent could assist regular users with tasks such as browsing and comparing products, managing purchases, assembling wish lists, or coordinating returns, while also helping them create and edit posts, reels, or stories for friends and family — not as professional content, but as casual, social, and entertainment-driven output.

That framing aligns closely with Meta’s historical strengths. The company has been most successful when AI capabilities are tightly integrated into high-frequency consumer workflows rather than positioned as standalone enterprise software.

A Manus-powered agent that helps users do things — shop, create, plan, or manage interactions inside Meta’s apps — would fit naturally into Instagram and Facebook’s evolution toward more agentic experiences. In that scenario, Manus functions less as an enterprise brand and more as an invisible execution layer, powering AI assistants that operate natively within Meta’s consumer ecosystem, where scale, engagement, and commerce already converge.

As a result, the acquisition’s clearest relevance is not whether enterprises should standardize on Manus, but that investments in internal agent frameworks, orchestration layers, and governance now appear increasingly well-justified — because that is precisely the layer large platforms are now willing to pay for.